[Bioclusters] Optimal database fragments in mpiBLAST

[Tim Cutts]

On 14-Nov-03, Malay Kumar Basu wrote:
> Hello Gurus:
> 
> Here is my cluster config:
> 
> Master - cpu 2 Xeon with hyperthreading 2 GB RAM
> 4 x nodes each - cpu 1 P4 hyperthreading 1 GB RAM
> 
> SGE recognizes total 12 cpus.
> 
> When hyperthreading on the whole setup can have 12 cpus, otherwise 6. 
> What should be the optimal BLAST database fragments for mpiBLAST?

As I understand it, you won't get superlinear speedup until the
fragments are small enough to fit in L2 cache, and that's only practical
if you have a very large number of nodes, and a very large set of
queries to run through it.

Otherwise, I'd set the size so that the chunk sits nicely in RAM and
there's still room for the dynamic expansion that BLAST will need -
something like 512MB chunks (or total_db_size/6 if that's smaller).

But then we don't use mpiBLAST much here at all, so you might get a much
better answer from someone more experienced than me.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group
Wellcome Trust Sanger Institute, Hinxton, Cambridge, CB10 1SA, UK